Tech News Weekly 432: How Wrong Are Google's AI Overviews?

Release Date: April 9, 2026
Host: Micah Sargent
Guests: Amanda Silberling (TechCrunch), Rod Pyle (Ad Astra magazine, This Week in Space), Scott Stein (CNET)

Episode Overview

This episode dives deep into three major stories transforming the tech landscape:

1. Google’s AI Overviews – How accurate are they really, and what does this mean for billions of users?
2. AI in Journalism – Should journalists use generative AI, and where do we draw the ethical line?
3. Artemis II and Apple’s Next 50 Years – Interviews exploring humanity’s return to the moon, and how Apple might evolve in decades to come.

Segment 1: How Reliable Are Google's AI Overviews?

Panel: Micah Sargent (Host), Amanda Silberling (TechCrunch)
Timestamps: 00:43–17:13

Key Discussion Points

• The New York Times Investigation

  • Google’s AI Overviews, now a staple atop most search results, were put under scrutiny using benchmarks by the Times and AI startup Umi.
  • Findings: 91% accuracy—improved from 85% with the Gemini 3 update—but with 5 trillion annual searches, a 9% error rate equates to tens of millions of potentially wrong answers every hour.
  • Critical Issue: Even "correct" overviews often cite sources that don’t actually support the answer (referred to as being "ungrounded").
    • In October, 37% of correct answers were ungrounded; by February, this rose to 56% after an AI upgrade.

• User Experience Anecdotes

  • Amanda describes using the AI Overview for a misspelled video game query, only to get a bizarre, incorrect medical-sounding response—a testament to persistent problems.
    • “Pikopoeia is the name of the sensation in which you suck up water and it gets lodged in your throat or like... I don't know what.” — Amanda Silberling (05:40)
  • Both Amanda and Micah admit to rarely engaging with the AI Overview, preferring to skip to original sources.
  • Older Millennials and Gen X seem more likely to trust or read overviews, while tech-savvy circles tend to scroll past.

• Statistical Perspective

  • The notion of "small error rates" becomes alarming at scale:
    • OpenAI’s own stat (0.15% of ChatGPT conversations about suicide) = over a million conversations, showcasing how “rounding error” becomes significant at size.
    • “Even if it's inaccurate like 0.1% of the time, the numbers are even bigger with Google than with ChatGPT...any tiny percent of error is so magnified and impacts so many millions of people in a way that is alarming.” — Amanda Silberling (12:31)

• Source Citation Problems

  • Google’s Overviews can provide correct information, but supporting links often don’t back up the details—creating a false sense of trustworthiness.
  • Example: The AI Overview got a baseball player’s age at death right, but linked to a source that gave the wrong death date.
  • “The fine print beneath every AI overview reads, ‘AI can make mistakes, so double check responses.’ But the fact that it's prominently placed at the top...doesn't encourage anyone to be skeptical.” — Micah Sargent (16:22)

Quotes & Memorable Moments

• "Why would I ever think that we should be thinking about people over profit? Not at all, sadly." — Micah (17:03)
• Humorous exchange about ideal disclaimers: “It should say, here's our AI overview. It might be wrong. So really, is this worth your time at all?” — Micah (16:59)

Segment 2: AI and the Future of Journalism

Panel: Micah Sargent, Amanda Silberling
Timestamps: 20:48–42:18

Key Discussion Points

• Investigative AI Use in Journalism

  • John Carreyrou (NYT) used AI to analyze decades of cryptography email lists, looking for linguistic fingerprints to analyze Satoshi Nakamoto’s identity.
  • Journalists increasingly use AI to summarize, draft, or analyze data—raising questions of appropriateness and transparency.
  • Amanda describes the divide:
    • Some independent journalists see LLMs as vital productivity tools for “one-person shows.”
    • But, major concerns persist about quality, hallucination risk, and authenticity.

• Ethics & Transparency

  • Micah underscores the importance of disclosure: “What I hate, though, is how often we hear after the fact that AI was used… and it is not always explicitly laid out and made clear. That frustrates me.”
  • Amanda compares AI use to Wikipedia: it may help expedite research, but shouldn’t be taken at face value or cited as a source.
    • “Maybe ChatGPT, like ... the main LLMs are sort of comparable to Wikipedia when you're in high school… if it is expediting the time you spend looking for the piece of information, I don't know if that is the worst thing for writers that are on tight deadlines.” — Amanda (29:10)
  • The “slippery slope” argument: Once AI is used for small editing or research, where does it stop?

• AI as a Research Tool

  • Both agree AI is best suited for synthesizing large datasets that humans couldn’t tackle alone.
  • Amanda points to positive uses—e.g., when AI helped analyze “Facebook files” whistleblower documents, it could have found important connections that manual sorting would miss.

• Skepticism & Risks

  • Both participants agree skepticism must remain high, especially since studies show up to 9% of news articles are partially or wholly AI-generated and half of AI news outputs have major integrity issues.

Notable Quotes

• “AI needs to be used in circumstances where if it fails, the worst thing that happens is that your cookies are crumbly.” — Amanda (36:27)
• “There are times where, if I want to know how much money Meta spent on Reality Labs in 2023, ChatGPT will probably give me an answer I can then fact check.” — Amanda (29:10)
• “AI is kind of like a teenager driving your Ferrari... the Ferrari isn't the problem, it's the person driving it who may not be skilled in driving a stick.” — Micah (32:26)
• On sustainability:
  • “Is it worth the cookies? To me, it's not worth the cookies.” — Micah (41:02)

Segment 3: Artemis 2—Humanity’s Return to the Moon

Interview: Micah Sargent with Rod Pyle (Ad Astra Magazine, This Week in Space)
Timestamps: 45:45–70:29

Key Discussion Points

• Artemis 2 Mission Overview

  • First crewed U.S. lunar flyby since 1972, echoing what Apollo 8 achieved—but with totally modern tech and goals.
  • “We haven’t left Earth orbit since 1972, which is a very long time. ...This mission was a bit of a follow to Apollo 8, if you will.” — Rod (46:12)
  • Objectives: Test Orion spacecraft systems, life support, and ground control for extended lunar missions.

• Mission Milestones

  • First crewed launch of SLS (Space Launch System).
  • Translunar injection burn—pushed the spacecraft on a free-return flyby loop around the moon.
  • Only minor course corrections needed; modern computing enabled tighter navigation than ever.

• Record Breaking and Emotional Moments

  • Furthest distance from Earth by humans (necessitated by mission's unique trajectory).
  • Crater on the far side named after commander’s late wife, Carol, bringing Mission Control and listeners to tears.
    • “We’d like to name this crater after Reed’s wife, Carol… That was really one. There was not a dry eye in the house.” — Rod (58:01)
  • Touching messages from Apollo astronauts, including a recorded greeting from Jim Lovell: “Welcome to my own neighborhood… Don’t forget to enjoy the view.”

• Science and Engineering Insights

  • Not major new moon discoveries—most attention was on engineering validation.
  • Crew observed meteorite impacts (“white flashes”) on the lunar surface.
  • Parachute and ablative heat shield performance remains a key concern for re-entry.

• The Road Ahead for Artemis

  • Future missions (Artemis 3 and beyond) will use lessons learned here to build toward a true lunar base.
  • Shift in plans: Artemis 3 will now be an orbital/docking mission due to lander delays, with a surface landing targeted for Artemis 4.
  • NASA intends to accelerate lunar presence due to competition with China's space program.

Quotes

• “Almost half of all AI-generated responses had at least one significant issue with news integrity... roughly 9% of newly published newspaper articles were either partially or fully AI generated.” — Micah (32:26)
• On the lunar base goal:
  • “Again, why are we doing this? We've been talking about it for an awful long time, ... but China said, hey, not only are we going to land people on the moon, we're going to build a lunar base…” — Rod (69:52)

Segment 4: Apple’s Next 50 Years—A Futurist Speculation

Interview: Micah Sargent with Scott Stein (CNET)
Timestamps: 74:05–101:34

Key Discussion Points

• Why Predict Apple's Future?

  • Inspired by Apple’s 50th anniversary, CNET asked Scott to run scenarios on what Apple will look like in another half-century.
  • “Fifty years is an absurdly long span… And for over a decade, I thought the iPad and Mac would merge, and they still haven’t fully done that. So maybe that’s the right timespan!” — Scott Stein (75:31)

• Premium Tech vs. Inequality

  • Will Apple serve the mainstream, or retreat even further into luxury as the wealth gap grows?
  • Futurist Annie Hardy points out the possibility of a cyberpunk scenario—Apple straddling both mass and elite markets.

• Spatial Computing, Cameras, and AI

  • Apple’s focus on spatial computing (Vision Pro, immersive cameras, sensors) could reshape how we record and relive memories.
    • Gaussian splats and 3D scanning as the next leap in consumer memory capture.
  • The lines blur between VR/AR, robotics, and contextual AI—all interconnected tech areas where Apple may forge ahead.

• The Persistent iPhone (and What Comes After)

  • “Laptops… I thought would be gone. No, they’re still really useful. Phones are so essential right now… I just don’t see how they will disappear.” — Scott Stein (85:46)
  • Phones may become hubs powering a web of wearables: glasses, rings, hearables, etc., but may look much different than today.

• Wearables, Health, and Aging

  • As populations age, Apple’s long-term survival may hinge on building trustworthy health tech for lifelong users (“intelligently designed smart care”).
  • Future Apple devices may leverage decades of user health data for personalized care.

• Will Apple Still Exist in 2076?

  • No guarantees, but size and ecosystem breadth suggest yes—if the company remains willing to disrupt itself.
  • “Their presence in services is interesting... their future is flexible.” — Scott Stein (96:52)
  • Anecdotal reminder: “I asked my 13-year-old if Apple would still be around: ‘They’ll be dead.’”

Notable Quotes & Moments (w/ Timestamps)

• "Millions of people are talking to ChatGPT about things that perhaps they should be talking to with a professional or a friend or someone who is not a computer." — Amanda Silberling (12:29)
• “Honestly, I will say, when I saw this piece on CNET, it was absolutely an insta-read for me. Scary stuff—it's a lot to try to process and think about." — Micah Sargent (74:55)
• "When I look at things like smart watches, ... at first no one’s doing it, then everybody’s doing it." — Scott Stein (98:08)
• “I want to believe that everybody would [catch an AI hallucination], but maybe there's someone out there who would go, well, it must know.” — Micah Sargent (36:41)

Further Resources & Where to Find Guests

• Amanda Silberling:
  • TechCrunch | Podcast: Wow, If True — Bluesky: @manda.omglol
• Rod Pyle:
  • National Space Society (nss.org), Ad Astra Magazine (adastramagazine.com), podcast: This Week in Space
• Scott Stein:
  • CNET (Scott's Profile), Newsletter: The Intertwix on Beehive, Bluesky: @scottstein

Summary: What To Take Away

Google’s AI Overviews are still prone to error and misleading citations—a small error rate at Google’s scale affects millions. AI is a powerful but ethically fraught tool for journalists—synthesizing complex data is its strength, but transparency and skepticism are essential. Artemis II marks a triumphant and emotional step back to lunar exploration, laying groundwork for the next era of space travel. Apple’s future may hinge on health tech, spatial computing, and adapting to serve both the elite and the mass market—even as generational attitudes shift and new categories supplant old ones.

For full context, skip the ads. For the future, keep your skepticism at the ready—and double check those AI overviews!